Struct ComputeGraph 

pub struct ComputeGraph {
    pub nodes: Vec<TensorNode>,
    pub n_leafs: usize,
}

Compute graph: topologically sorted list of tensor operations.

Built once per model architecture, reused every decode step. Only the data pointers and parameters change between steps.

Fields

nodes: Vec<TensorNode>

Nodes in topological order (leafs first, output last)

n_leafs: usize

Number of leaf nodes (inputs, no operation)

Implementations

impl ComputeGraph

pub fn new() -> Self

Create an empty compute graph.

pub fn add_leaf(&mut self, data_ptr: u64, shape: [u32; 4]) -> usize

Add a leaf node (input tensor, no operation).

pub fn add_op( &mut self, op: TensorOp, data_ptr: u64, shape: [u32; 4], inputs: Vec<usize>, params: OpParams, ) -> usize

Add an operation node with inputs.

pub fn n_ops(&self) -> usize

Number of operation nodes (excludes leafs).

Trait Implementations

impl Clone for ComputeGraph

fn clone(&self) -> ComputeGraph

Returns a duplicate of the value.

fn clone_from(&mut self, source: &Self)

Performs copy-assignment from source.

impl Debug for ComputeGraph

fn fmt(&self, f: &mut Formatter<'_>) -> Result

Formats the value using the given formatter.

impl Default for ComputeGraph

fn default() -> Self

Returns the “default value” for a type.